Workflows

Community

Summit

2024

Future Trends and

Challenges in

Scientific

Workflows

September 3–5, 2024

Time- Sensitive

Workflows

Convergence of AI

and HPC Workflows

Multi- Facility

Workflows in

Next- Generation

Scientific

Collaboration

Heterogeneous HPC

Environments

User Experience and

Interfaces

FAIR Computational

Workflows

ORNL/TM-2024/3573

WORKFLOWS COMMUNITY SUMMIT 2024

Disclaimer. This research used resources of the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory,

which is supported by the Ofﬁce of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.

This report was prepared as an account of work sponsored by agencies of the United States Government. Neither the United

States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any

legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process

disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any speciﬁc commercial

product, process, or service by trade name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply its

endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of

authors expressed herein do not necessarily state or reﬂect those of the United States Government or any agency thereof.

License. This report is made available under a Creative Commons Attribution 4.0 International Public license (https://

creativecommons.org/licenses/by/4.0).

Preferred citation

R. Ferreira da Silva, D. Bard, K. Chard, S. DeWitt, I. T. Foster, T. Gibbs, C. Goble, W. Godoy, J. Gustafsson, U.-U. Haus, S.

Hudson, S. Jha, L. Los, D. Paine, F. Suter, L.Ward, S. Wilkinson, M. Amaris, Y. Babuji, J. Bader, R. Balin, D. Balouek, S.

Beecroft, K. Belhajjame, R. Bhattarai, W. Brewer, P. Brunk, S. Caino-Lores, H. Casanova, D. Cassol, J. Coleman, T. Coleman, I.

Colonnelli, A. A. Da Silva, D. de Oliveira, P. Elahi, N. Elfaramawy, W. Elwasif, B. Etz, T. Fahringer, W. Ferreira, R. Filgueira, J.

Fosso Tande, L. Gadelha, A. Gallo, D. Garijo, Y. Georgiou, P. Gritsch, P. Grubel, A. Gueroudji, Q. Guilloteau, C. Hamalainen,

R. Hong Enriquez, L. Huet, K. Hunter Kesling, P. Iborra, S. Jahangiri, J. Janssen, J. Jordan, S. Kanwal, L. Kunstmann, F.

Lehmann, U. Leser, C. Li, P. Liu, J. Luettgau, R. Lupat, J. M. Fernandez, K. Maheshwari, T. Malik, J. Marquez, M. Matsuda, D.

Medic, S. Mohammadi, A. Mulone, J. Navarro, K. Ng, K. Noelp, B. P. Kinoshita, R. Prout, M. R. Crusoe, S. Ristov, S. Robila,

D. Rosendo, B. Rowell, J. Rybicki, H. Sanchez, N. Saurabh, S. Saurav, T. Scogland, D. Senanayake, W. Shin, R. Sirvent, T.

Skluzacek, B. Sly-Delgado, S. Soiland-Reyes, A. Souza, R. Souza, D. Talia, N. Tallent, L. Thamsen, M. Titov, B. Tovar, K.

Vahi, E. Vardar-Irrgang, E. Vartina, Y. Wang, M. Wouters, Q. Yu, Z. Al Bkhetan, M. Zulﬁqar, “Workﬂows Community Summit

2024: Future Trends and Challenges in Scientiﬁc Workﬂows”, Technical Report, ORNL/TM-2024/3573, October 2024, DOI:

10.5281/zenodo.13844759.

@techreport{wcs2024,

author = {Ferreira da Silva, Rafael and Bard, Deborah and Chard, Kyle and DeWitt, Shaun and Foster, Ian T. and Gibbs, Tom and Goble,

Carole and Godoy, William and Gustafsson, Johan and Haus, Utz−Uwe and Hudson, Stephen and Jha, Shantenu and Los,

Laila and Paine, Drew and Suter, Frederic and Ward, Logan and Wilkinson, Sean and Amaris, Marcos and Babuji, Yadu and

Bader, Jonathan and Balin, Riccardo and Balouek, Daniel and Beecroft, Sarah and Belhajjame, Khalid and Bhattarai, Rajat and

Brewer, Wes and Brunk, Paul and Caino−Lores, Silvina and Casanova, Henri and Cassol, Daniela and Coleman, Jared and

Coleman, Taina and Colonnelli, Iacopo and Da Silva, Anderson Andrei and de Oliveira, Daniel and Elahi, Pascal and

Elfaramawy, Nour and Elwasif, Wael and Etz, Brian and Fahringer, Thomas and Ferreira, Wesley and Filgueira, Rosa and Fosso

Tande, Jacob and Gadelha, Luiz and Gallo, Andy and Garijo, Daniel and Georgiou, Yiannis and Gritsch, Philipp and Grubel,

Patricia and Gueroudji, Amal and Guilloteau, Quentin and Hamalainen, Carlo and Hong Enriquez, Rolando and Huet, Lauren

and Hunter Kesling, Kevin and Iborra, Paula and Jahangiri, Shiva and Janssen, Jan and Jordan, Joe and Kanwal, Sehrish and

Kunstmann, Liliane and Lehmann, Fabian and Leser, Ulf and Li, Chen and Liu, Peini and Luettgau, Jakob and Lupat, Richard

and Fernandez, Jose M. and Maheshwari, Ketan and Malik, Tanu and Marquez, Jack and Matsuda, Motohiko and Medic,

Doriana and Mohammadi, Somayeh and Mulone, Alberto and Navarro, John−Luke and Ng, Kin Wai and Noelp, Klaus and

P. Kinoshita, Bruno and Prout, Ryan and R. Crusoe, Michael and Ristov, Sashko and Robila, Stefan and Rosendo, Daniel and

Rowell, Billy and Rybicki, Jedrzej and Sanchez, Hector and Saurabh, Nishant and Saurav, Sumit Kumar and Scogland, Tom and

Senanayake, Dinindu and Shin, Woong and Sirvent, Raul and Skluzacek, Tyler and Sly−Delgado, Barry and Soiland−Reyes,

Stian and Souza, Abel and Souza, Renan and Talia, Domenico and Tallent, Nathan and Thamsen, Lauritz and Titov, Mikhail and

Tovar, Ben and Vahi, Karan and Vardar−Irrgang, Eric and Vartina, Edite and Wang, Yuandou and Wouters, Merridee and Yu, Qi

and Al Bkhetan, Ziad and Zulﬁqar, Mahnoor},

title = {{Workﬂows Community Summit 2024: Future Trends and Challenges in Scientiﬁc Workﬂows}},

year = {2024},

publisher = {Zenodo},

number = {ORNL/TM−2024/3573},

doi = {10.5281/zenodo.13844759},

institution = {Oak Ridge National Laboratory}

}

WORKFLOWS COMMUNITY SUMMIT 2024

Table of Contents

Executive Summary 4

1 Introduction 5

1.1 The Workﬂows Community Summit 2024 . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2 Summit Structure and Activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 A Look Back at Workﬂows Community Summits 7

3 Time-Sensitive Workﬂows 8

4 Convergence of AI and HPC Workﬂows 10

5 Multi-Facility Workﬂows in Next-Generation Scientiﬁc Collaboration 11

6 Heterogeneous HPC Environments 13

7 User Experience and Interfaces 15

8 FAIR Computational Workﬂows 16

References 19

Appendix A: Participants and Contributors 23

Appendix B: Agenda 25

WORKFLOWS COMMUNITY SUMMIT 2024

Executive Summary

KEY RECOMMENDATIONS

Time Sensitive: Develop standardized cross-facility au-

thentication and data protocols to enable secure, near

real-time data movement and analysis across distributed

research infrastructures.

AI-HPC: Invest in R&D of AI-driven workﬂow opti-

mization techniques to automate the design and execu-

tion of complex HPC workﬂows.

Multi-Facility: Implement distributed-by-design work-

ﬂow models to create workﬂows that inherently account

for multi-facility execution, incorporating environment

descriptions and constraints.

Heterogeneous Environments: Develop a community-

driven suite of diverse workﬂow benchmarks to guide

system evaluation, procurement, and performance opti-

mization across heterogeneous HPC environments.

UX and Interfaces: Formulate and implement a set of

UX principles speciﬁcally tailored for scientiﬁc work-

ﬂows to guide the design of more intuitive, user-friendly

interfaces across multiple workﬂow systems.

FAIR: Create a FAIR workﬂow maturity model and as-

sociated metrics to guide the implementation of FAIR

principles and provide a standardized way to assess and

improve workﬂow FAIRness.

The Workﬂows Community Summit con-

vened 111 participants from 18 countries to ex-

plore emerging trends and challenges in scien-

tiﬁc workﬂows. It focused on six key areas that

are shaping the future of computational research:

time-sensitive workﬂows, the convergence of

AI and HPC, multi-facility workﬂows, heteroge-

neous HPC environments, user experience and

interfaces, and FAIR computational workﬂows.

The integration of AI and exascale com-

puting has revolutionized scientiﬁc workﬂows,

enabling higher-ﬁdelity models and more com-

plex, time-sensitive processes. However, this

advancement introduces signiﬁcant challenges

in managing heterogeneous environments and

multi-facility data dependencies. To address

these issues, innovative approaches like trust

bundles and advanced data fabric solutions are

being developed. Simultaneously, the rise of

large language models is pushing computational

demands to zettaﬂop scales, driving the devel-

opment of modular, adaptable systems and cloud-service models. These new paradigms aim to optimize

resource utilization, enhance scalability, and ensure reproducibility across diverse computing infrastructures.

As data volumes and velocity continue to grow, there is an increasing demand for workﬂows that

span multiple facilities. This shift presents challenges in data movement, curation of large-scale data,

and overcoming institutional silos. The increasing prevalence of diverse hardware architectures, including

specialized AI accelerators, requires a focus on integrating workﬂow considerations into early system design

stages and developing standardized resource and data management tools. The summit also emphasized the

critical importance of user experience in workﬂow systems, with on-going efforts to simplify workﬂow usage,

improve graphical representations, and adapt concepts from commercial software development to scientiﬁc

applications. Additionally, ensuring Findable, Accessible, Interoperable, and Reusable (FAIR) workﬂows is

crucial for enhancing collaboration and accelerating scientiﬁc discovery, but challenges remain in balancing

standardization with innovation and applying FAIR principles to workﬂows with conﬁdential components.

Key recommendations include developing standardized metrics for time-sensitive workﬂows, creating

frameworks for cloud-HPC integration, implementing distributed-by-design workﬂow modeling, establishing

multi-facility authentication protocols, and accelerating AI integration in HPC workﬂow management. The

summit also emphasized the need for comprehensive workﬂow benchmarks, formulating workﬂow-speciﬁc

UX principles, and creating a FAIR workﬂow maturity model. These recommendations underscore the

importance of continued collaboration and innovation in addressing the complex challenges posed by the

convergence of AI, HPC, and multi-facility research environments.

WORKFLOWS COMMUNITY SUMMIT 2024

1 Introduction

The landscape of scientiﬁc workﬂows is undergoing a profound transformation, driven by the convergence

of high performance computing (HPC) and artiﬁcial intelligence (AI) [

;

]. Scientiﬁc workﬂows are

evolving from being mere tools to becoming the new applications that drive forward the frontiers of science.

This evolution is characterized by increasing complexity, data volume, and computational demands, thus

necessitating robust, adaptable, and ﬂexible support systems. Key trends that shape the future of scientiﬁc

workﬂows include the deep integration of AI with HPC, the rise of multi-facility workﬂows spanning multiple

research institutions, the shift towards data-driven workﬂow dynamics, the challenges posed by heterogeneous

computing environments, and the pressing need for sustainable, energy-efﬁcient practices [

;

]. These

developments present both exciting opportunities and signiﬁcant challenges for the community, requiring

innovative approaches to workﬂow design, management, and execution. As we enter the postexascale era,

the demand for processing scientiﬁc data continues to grow, alongside steady enhancements in large-scale

HPC capabilities. The integration of edge computing with traditional HPC resources is creating a more

ﬂuid and efﬁcient computing continuum [

], further complicating the workﬂow landscape. Additionally, the

need for near real-time analysis capabilities and improved workﬂow resilience is driving the development of

more dynamic and ﬂexible data management solutions. The Workﬂows Community Summit aims to address

these emerging trends and challenges, fostering collaboration and knowledge exchange to shape the future of

scientiﬁc workﬂows in an era of unprecedented technological advancement and environmental awareness.

1.1 The Workﬂows Community Summit 2024

This document presents the outcomes and insights from the 2024 international edition of the “Workﬂows

Community Summit,” held from September 3–5, 2024 [

]. The three-day summit brought together a diverse

group of 111 participants (Figure 1), representing a global community of researchers, developers, and

industry professionals. The event expanded for the ﬁrst time, adding an extra day with a timezone favorable

for Oceania and Asia. Attendees represented 18 countries across six continents, including Australia, Austria,

Brazil, France, Germany, India, Italy, Japan, Latvia, New Zealand, Puerto Rico, Singapore, Spain, Sweden,

Switzerland, The Netherlands, United Kingdom, and the United States. This international cohort comprised

experts from various workﬂow management systems (WMSs), end-users, and industry representatives,

fostering a rich environment for cross-cultural and interdisciplinary exchange. The summit’s increasing

global reach underscores the increasing importance of scientiﬁc workﬂows in addressing complex, large-scale

research challenges that transcend national boundaries. In future years, the event will seek to expand its reach

further, inviting contributions from workﬂow experts across domains and disciplines.

The summit was organized by key members of the Workﬂows Community Initiative (WCI) [

], a

volunteer-driven organization dedicated to unifying the diverse workﬂows community. The WCI’s mission

is to provide valuable community resources and capabilities that empower scientists and workﬂow system

developers. The initiative serves as a catalyst for community-wide efforts to address the grand challenges

facing scientiﬁc workﬂows. By facilitating this summit, the WCI continues its commitment to advancing the

state-of-the-art in workﬂow technologies and methodologies, ultimately accelerating scientiﬁc discovery and

innovation across disciplines.

1.2 Summit Structure and Activities

Building on trends discussed in [

], the six cross-cutting summit’s topics reﬂect emerging challenges and

opportunities in the rapidly evolving landscape of scientiﬁc computing (Table 1): (i) “Convergence of AI and

WORKFLOWS COMMUNITY SUMMIT 2024

Figure 1: Participants of the 2024 Workﬂows Community Summit (virtual, Sept. 3–5, 2024).

HPC Workﬂows” addresses the increasing synergy between AI systems and traditional HPC simulations;

(ii) “Multi-Facility Workﬂows in Next-Generation Scientiﬁc Collaboration” explores the complexities of

orchestrating workﬂows across diverse research infrastructures; (iii) “Time-Sensitive Workﬂows” focuses on

near real-time data integration and processing challenges; (iv) “Heterogeneous HPC Environments” tackles

the need to navigate performance variability across diverse hardware conﬁgurations; (v) “User Experience and

Interfaces” recognizes the growing importance of accessibility and usability in complex scientiﬁc workﬂows;

and (vi) “FAIR Computational Workﬂows” emphasizes the need for ﬁndable, accessible, interoperable, and

reusable workﬂow practices. Collectively, these topics address critical trends and challenges identiﬁed in the

previous years’ summits, aiming to shape the future of scientiﬁc workﬂows in the context of rapidly evolving

HPC technologies and practices.

For each of the technical topics listed in Table 1, the co-leaders delivered concise 10-minute lightning

talks in plenary sessions, followed by in-depth discussions in targeted breakout groups. These sessions

had dual objectives: (i) to evaluate previously identiﬁed challenges and their proposed solutions, and (ii) to

pinpoint critical gaps and formulate potential short- and long-term strategies for supporting emerging and

WORKFLOWS COMMUNITY SUMMIT 2024

Topic Discussion Co-Leaders

Topic 1: Time-Sensitive Workﬂows Utz-Uwe Haus (HPE), Shaun DeWitt (UKAEA)

Topic 2: Convergence of AI and HPC Workﬂows Shantenu Jha (PPPL), Logan Ward (ANL)

Topic 3: Multi-Facility Workﬂows in Next-Generation

Scientiﬁc Collaboration

Tom Gibbs (NVIDIA), Debbie Bard (LBNL)

Topic 4: Heterogeneous HPC Environments William Godoy (ORNL), Stephen Hudson (ANL)

Topic 5: User Experience and Interfaces Drew Paine (LBNL), Laila Los (UFreiburg)

Topic 6: FAIR Computational Workﬂows Carole Goble (UManchester), Sean Wilkinson (ORNL)

Spotlight Discussion: What makes a workﬂow good?

Johan Gustafsson (Australian BioCommons, University of Melbourne)

Table 1: Workﬂows Community Summit topics and discussion co-leaders.

novel workﬂow applications in light of the rapidly evolving computing continuum paradigm. Day 3 of the

event featured a session of lightning talks highlighting workﬂow-related activities from Oceania and Asia.

All presentations and videos can be found in the summit website (https://workﬂows.community/

summits/2024), and videos can be watched from WCI’s YouTube channel (https://www.youtube.com/

watch?v=STy5HwEVj8k&list=PLAtmuqHExRvOzPOvJfSH8PwjbVJFyulCy).

Keynote Speaker. The summit was further enriched by an invited keynote address from Ian T. Foster

(Argonne National Laboratory / University of Chicago), a distinguished leader in Scientiﬁc Computing,

who spoke on “AI-mediated Scientiﬁc Workﬂows.” Dr. Foster presented a vision of accelerated scientiﬁc

discovery through the integration of AI methods into various stages of the scientiﬁc process. He discussed

the development of AI-driven “embodied agents” capable of performing tasks such as literature review,

hypothesis generation, and experimental design. Foster highlighted a speciﬁc example in antimicrobial

peptide discovery, where AI agents can interact with databases, run simulations, and even control robotic

laboratories. He emphasized the need for specialized language models designed for scientiﬁc applications,

and the importance of creating persistent, self-improving AI systems. Foster also addressed the computational

challenges of these AI-mediated workﬂows, including the need for secure, reliable, and scalable infrastructure

to support long-running, stateful processes across diverse computing environments.

Spotlight Discussion. The spotlight discussion “What makes a workﬂow good?”, led by Johan Gustafsson

from Australian BioCommons [

], explored the intricacies of workﬂow quality in scientiﬁc computing.

Participants addressed the tension between research outcomes and the demand for accessible, reproducible

workﬂows. Key attributes of exemplary workﬂows include containerization for stability, rigorous validation,

comprehensive documentation, and ongoing support. Standard software practices, especially unit testing,

and the role of Research Software Engineers in developing and maintaining high-quality workﬂows were

emphasized for reliability. Challenges discussed included balancing user-friendliness with preventing misuse,

and managing diverse expertise across the workﬂow lifecycle. A “good” workﬂow was deﬁned as robust,

purpose-driven, environment-agnostic, and well-documented. This holistic approach to workﬂow quality

reﬂects the evolving scientiﬁc computing landscape, demonstrating a shared commitment to sustainable,

reproducible methodologies that advance scientiﬁc discovery.

2 A Look Back at Workﬂows Community Summits

The workﬂows community took a signiﬁcant step forward with the publication of “A Community

Roadmap for Scientiﬁc Workﬂows Research and Development” [

] in 2021. This pioneering paper, resulting

WORKFLOWS COMMUNITY SUMMIT 2024

from the ﬁrst series of Workﬂows Community Summits, presented a consolidated view of the state of the art,

challenges, and potential future directions in the ﬁeld of scientiﬁc workﬂows. The roadmap identiﬁed six

key themes: FAIR computational workﬂows, AI workﬂows, exascale challenges, APIs and interoperability,

training and education, and building a workﬂows community. For each theme, the roadmap outlined speciﬁc

challenges and proposed community activities to address them. Additionally, it provided a set of technical

milestones across three main thrusts: deﬁning common workﬂow patterns and benchmarks, identifying paths

toward interoperability, and improving workﬂow systems’ interface with HPC software and hardware stacks.

Building on this foundation, the Workﬂows Community Summit 2022 further advanced the discourse on

scientiﬁc workﬂows. The summit report [

] expanded on the previous themes and introduced new focus

areas: speciﬁcations, standards, and APIs; AI workﬂows; high-performance data management and in situ

workﬂows; HPC and quantum workﬂows; FAIR computational workﬂows; and workﬂows for continuum

and cross-facility computing. Each topic was thoroughly discussed, with participants identifying current

challenges and proposing both short- and long-term solutions. The summit highlighted the evolving landscape

of scientiﬁc computing and the need for WMSs to adapt to emerging paradigms such as edge-to-HPC

computational workﬂows and the integration of quantum computing with classical HPC systems.

Most recently, the paper “Frontiers in Scientiﬁc Workﬂows: Pervasive Integration with High-Performance

Computing” [

] provides a forward-looking perspective on the ﬁeld by identifying ﬁve key trends shaping the

future of workﬂows: increased synergy between AI and HPC workﬂows, the rise of cross-facility workﬂows,

data-driven HPC workﬂow dynamics, managing performance variability in heterogeneous HPC environments,

and the push for sustainable practices in HPC workﬂow design. For each trend, the paper predicts future

developments: AI systems for science trained on HPC workﬂow data, global interoperability standards

for multi-facility collaboration, dynamic near real-time workﬂow control planes, software solutions for

consistent performance across heterogeneous hardware, and AI-driven optimization of HPC workﬂows for

energy efﬁciency and reduced environmental impact. It emphasizes the ﬁeld’s rapid evolution and scientiﬁc

workﬂows’ key role in advancing cross-disciplinary research.

3 Time-Sensitive Workﬂows

Time-sensitive workﬂows have been integral to a diverse range of scientiﬁc and technological domains,

from autonomous vehicles and aircraft control systems to cutting-edge research in beamline-based materials

science, astronomy and astrophysics, observational science, and experimental fusion [

]. These workﬂows

are characterized by their urgent nature, requiring real-time or near real-time responses across multiple

facilities or resources to enable timely decision-making, experiment steering, and data analysis [12].

Emerging Paradigms and Challenges. Since the 2022 summit, signiﬁcant changes have occurred in this

ﬁeld, driven by advancements in AI, machine learning, and exascale computing. The rise of AI and machine

learning has led to the increased use of surrogate models, replacing traditional reduced-order models due to

their speed and efﬁciency in time-critical situations. This shift has been particularly noticeable in autonomous

vehicles and other systems requiring rapid decision-making. Exascale computing has enabled larger models

with higher ﬁdelity and smaller time steps, pushing the boundaries of what is possible in computational

science. Additionally, the volume of data generated by exascale systems, such as the petabyte-per-day output

from projects like Destination Earth [

], poses signiﬁcant data movement and management challenges.

The integration of digital twins, the need for federation across facilities, and the growing importance of

provenance in regulated environments have added new layers of complexity. These advancements have

brought forth challenges in managing heterogeneous computing environments, dealing with cross-facility

WORKFLOWS COMMUNITY SUMMIT 2024

data dependencies, and addressing on-demand access and authentication issues in time-critical scenarios.

Recent Advances and Emerging Solutions. To address these challenges, several state-of-the-art approaches

are being explored. For on-demand access, attested execution with trust bundles is being considered, along

with federated identity and access management to overcome multi-factor authentication hurdles in urgent

computing scenarios. Data movement solutions are being developed both for high-performance workﬂows

within systems and for efﬁcient cross-site transfers. Provenance tracking for AI models and object stores is

becoming increasingly important, with potential solutions including the use of version control systems like

Git or frameworks such as the HDF5 Common Metadata Framework [

], and the adoption of standards like

RO-Crates [

] for workﬂow provenance in major life science platforms. The FAIR4ML [

] initiative is

proposing metadata schemas based on Schema.org and HuggingFace for capturing AI model descriptions.

For data access policies, research is ongoing to develop methods that can guarantee secure access to object

stores while complying with varying geographic regulations. The use of fabric-attached memory is being

explored for HPC, although it presents new security challenges beyond traditional POSIX-style models.

Discussion. Time-sensitive workﬂows are a key focus in scientiﬁc computing, driving efforts to reﬁne

deﬁnitions and research priorities. Emphasis is on distinguishing between “time-critical” (real-time) and

“time-sensitive” (workﬂow triggering) processes, especially for near real-time control scenarios, such as

device output to input parameters with time window constraints. The US Department of Energy (DOE)

Ofﬁce of Science’s deﬁnition of time-sensitive workﬂows as “low-latency workﬂows requiring real-time, or

near-real-time, response across more than one facility or resource for timely decision-making and experiment

steering” is recognized, emphasizing these workﬂows’ multi-facility nature. A meta-analysis of 74 DOE

Ofﬁce of Science case studies found 45% of workﬂows to be time-sensitive [17].

Addressing time-sensitive workﬂow challenges involves key-value stores for smart applications, self-

describing datasets for partial reads, and access pattern-based data migration. The expanding computing

continuum (HPC, HTC, cloud, edge) necessitates advanced data movement and workﬂow management.

This has led to interest in hierarchical WMSs, with cloud orchestrators coordinating specialized workﬂow

managers for ﬁle-based, stream-focused, and other specialized tasks. Efforts focus on unifying WMSs

to improve interoperability and reduce custom middleware. Robust provenance checking is emphasized,

especially for long-term data and model tracking in regulated industries like fusion energy and life sciences.

Data access policies in Trusted Research Environments (TREs) [

] in health data research are a key

focus, with emphasis on controlled code movement and frameworks such as GA4GH passports [

] and

the Five Safes model [

]—they ensure data security and regulatory compliance across diverse geographic

contexts. Trusted Execution Environments (TEEs), established by hardware extensions (e.g., Intel

SGX,

Intel

TDX, or AMD

SEV), are being explored for end-to-end distributed workﬂow conﬁdentiality [

;

Integrating AI and machine learning models into heavily regulated environments, such as power plant control

systems, raises concerns about model construction, validation, and metadata management.

The potential of serverless workﬂows in time-sensitive applications present both scalability advantages and

data locality challenges. The scientiﬁc community has increased its efforts to enhance workﬂow portability

across diverse computing environments, focusing on innovative solutions that leverage containerization

technologies, virtual machines, and advanced scheduler integration techniques. These approaches aim to

create more ﬂexible, adaptable, and efﬁcient WMSs capable of seamlessly operating across heterogeneous

infrastructures. Initiatives like Autosubmit [

], which demonstrates portability across diverse systems

including Fugaku, exemplify promising advances in workﬂow interoperability.

WORKFLOWS COMMUNITY SUMMIT 2024

Recommendations:

•

Develop standardized deﬁnitions and metrics for time-sensitivity in workﬂows to facilitate

better communication and comparison across different domains and applications.

•

Invest in middleware solutions that can efﬁciently manage data movement and access across

heterogeneous computing environments, focusing on both performance and security aspects.

•

Establish best practices and frameworks for provenance tracking in AI and machine learning

workﬂows, with emphasis on regulated environments and long-term data retention requirements.

•

Unite workﬂow system developers, cloud service providers, and HPC centers to create portable

solutions that can seamlessly operate across computing paradigms.

4 Convergence of AI and HPC Workﬂows

The integration of AI and HPC represents a paradigm shift in scientiﬁc and computational capabilities [

This convergence is multiplicative in its impact, enabling complex simulations and data-driven discoveries at

unprecedented scales. AI-powered HPC workﬂows are drastically booting research efﬁciency across diverse

scientiﬁc domains [

;

]. While massive datasets generated by HPC simulations fuel the development and

training of next-generation models implemented in AI systems, these AI systems, in turn, dynamically steer

simulations for optimizing resource utilization and for maximizing scientiﬁc output.

Emerging Paradigms and Challenges. Large language models (LLMs) have become dominant, requiring

exascale resources and motivating the development of workﬂows can perform zettaﬂops-scale computation.

This shift has occurred alongside the continued growth of smaller-scale AI applications, including AI-driven

robotics integrated with HPC workﬂows. The diversiﬁcation of AI applications has expanded beyond proof-

of-concept demonstrations to become integral components of larger frameworks and end-goals in themselves.

However, persistent challenges remain. These include optimizing resource utilization, managing workﬂow

performance and latency, and effectively routing data for inference tasks. Additionally, the community faces

new hurdles in environment management, with the fast-paced AI ecosystem often outstripping traditional

HPC build environments. Elasticity in compute and data capacity for AI training, along with reproducibility,

reusability, and explainability of AI workﬂows on HPC systems, have emerged as critical concerns.

Recent Advances and Emerging Solutions. To address these challenges, the community is enhancing existing

workﬂows and developing decoupled data fabrics [

], with implementations based on ADIOS2 [

] (e.g.,

INSTANT [

]), HDF5 [

] (e.g., Wilkins [

]), or pure POSIX semantics (e.g., CAPIO [

]). This approach

allows for greater ﬂexibility in interchanging components to optimize workﬂows under varying conditions.

AI startups have contributed to this trend by offering solutions that treat AI as distinct, interchangeable

components within larger workﬂows. The focus has shifted towards creating modular, adaptable systems that

can leverage the rapid advancements in AI techniques while maintaining the robustness required for HPC

applications. Containerization in HPC is experiencing continued development, which promises to alleviate

some of the environmental management issues. Cloud-service-oriented models are also being explored for

their potential to mesh well with workﬂows, especially in multi-site scenarios. Despite the potential beneﬁts,

the adoption of cloud services in academic and national laboratory settings is tempered by concerns over cost

structures, data governance, and the need to align with existing resource allocation models.

Discussion. The integration of AI into HPC workﬂows has gained increased attention within the scientiﬁc

community, centering on four key areas: cloud computing integration, provenance tracking, benchmark

WORKFLOWS COMMUNITY SUMMIT 2024

development, and AI-driven workﬂow optimization. While cloud services offer advantages such as elasticity

and access to specialized hardware, concerns persist regarding cost models, data locality, and the blurring of

boundaries between traditional HPC and cloud paradigms. The community recognizes the potential of cloud

integration, particularly for speciﬁc workﬂow phases like post-processing or components requiring dynamic

resource allocation, but acknowledges the need for careful consideration of implementation strategies in

academic and national laboratory settings.

Provenance tracking in AI workﬂows has emerged as another key area of discussion. As AI models grow

in complexity and are increasingly applied to critical decision-making processes, the ability to trace and

explain model predictions has become paramount [

]. This need for trustworthy AI models, i.e., models that

achieve a desirable balance between performance and principles like explainability, transparency, auditability,

accountability, and reproducibility extends beyond the training phase to encompass the entire lifecycle of AI

workﬂows, including the production and preparation of training data. The community faces the challenge of

implementing comprehensive provenance tracking without introducing signiﬁcant computational overhead,

underscoring the need for efﬁcient and standardized workﬂow provenance methodologies [32].

The development of benchmarks for AI-HPC workﬂows and the potential of AI-driven workﬂow optimiza-

tion have also garnered signiﬁcant attention. While the community agrees on the importance of benchmarks

for understanding performance bottlenecks and guiding hardware decisions, the diversity and rapid evolution

of AI applications present challenges in establishing representative workﬂow patterns. Concurrently, the

concept of “AI for Workﬂows” is opening new avenues for automating the design and execution of complex

HPC workﬂows. This paradigm shift raises intriguing questions about the optimal modularization of workﬂow

components to facilitate AI-driven learning and composition.

Recommendations:

•

Develop frameworks for integrating cloud services with HPC workﬂows, addressing cost models,

data lifecycle, and resource allocation to enable seamless hybrid cloud-HPC environments.

•

Establish comprehensive provenance tracking for AI workﬂows, balancing thoroughness with

efﬁciency to support explainability and reproducibility without sacriﬁcing performance.

•

Develop collaborative benchmarks representative of diverse AI-HPC workﬂow patterns across

scientiﬁc domains, partnering domain experts with computer scientists.

•

Invest in R&D of AI-driven workﬂow optimization techniques, focusing on modular workﬂow

designs that facilitate automated composition and adaptation of complex HPC workﬂows.

•

Examine key requirements for parallel/distributed workﬂow patterns and models suited to

large-scale AI/machine learning applications in science and engineering.

5 Multi-Facility Workﬂows in Next-Generation Scientiﬁc Collaboration

Advanced instruments now produce vast amounts of complex scientiﬁc data. As individual experiment

sites and computing facilities struggle to keep pace with unprecedented data rates and volumes, there is a

growing recognition of the need for multi-facility (or federated) workﬂows [

]. AI-enhanced multi-facility

workﬂows integrate capabilities across HPC and experimental and observational facilities [

;

], offering

near real-time data analysis, improved experimental resilience, and better resource utilization.

Emerging Paradigms and Challenges. The potential of multi-facility integrations has been recognized

at the highest levels, with multiple ﬁnalists and winners of the Gordon Bell Special Prize showcasing

WORKFLOWS COMMUNITY SUMMIT 2024

composite workﬂows with downstream connections to user facilities. However, this emerging paradigm

faces numerous challenges. These include multi-facility data movement, the need for curated large-scale

data, and the persistent silos between science facilities. The heterogeneous hardware and software across

HPC systems necessitate sophisticated, AI-enhanced middleware solutions for seamless workﬂow execution.

Efﬁcient data management across distributed environments requires robust transfer, storage, and security

strategies. Additional hurdles include network constraints, batch-oriented HPC facilities with limited

interactive capabilities, and project-based funding models that may not align with multi-facility collaborative

efforts. Moreover, scheduling and resource allocation across independent facilities pose signiﬁcant challenges.

Recent Advances and Emerging Solutions. The DOE’s Integrated Research Infrastructure (IRI) program [

]

leads these efforts with domain-speciﬁc pathﬁnder projects, while the ﬁeld sees the emergence of both

single-vendor ecosystems and cooperative system software stacks. The Superfacility [

] project integrates

experimental and computational resources across diverse scientiﬁc domains, enabling near real-time data

analysis. In astronomy, the Dark Energy Camera (DECam) Alliance leverages AI algorithms to manage data

across supercomputers and remote databases for identifying transient objects [

]. Despite these successes,

there is a growing need for more fundamental, systemic solutions to support complex workﬂows across

multiple facilities. Future research directions include exploring new scientiﬁc challenges suited to multi-

facility resources, understanding AI workﬂow requirements, addressing AI at the edge, integrating quantum

computing, and developing function-as-a-service capabilities.

Discussion. Multi-facility workﬂows present a complex array of challenges and opportunities, mirroring

the evolving landscape of large-scale scientiﬁc research. A key issue is the preparation of data for AI

applications. Participants emphasized that much experimental data is “dirty” – unlabeled and noisy – creating

signiﬁcant hurdles beyond conventional HPC and workﬂow concerns. This challenge is compounded by the

complexities of cross-facility data transfer and user interaction, especially when navigating the varied internal

policies and infrastructures of different facilities. The transition from individual user accounts to service

accounts introduces additional layers of complexity in operational and security aspects. This shift underscores

the urgent need for a standardized, robust authentication method that surpasses the security of current

solutions like SSH, while maintaining ease of use, performance, and interoperability across diverse systems.

Furthermore, workﬂow management across facilities presents its own set of unique challenges, including

the non-trivial task of migrating workﬂows between machines with different architectures and policies (see

Section 6). Participants emphasized the critical importance of identifying and categorizing various interaction

types and coupling modes within multi-facility hybrid workﬂows. A key step forward in addressing these

challenges is enabling users to model distributed-by-design workﬂows. Such workﬂows would incorporate

detailed descriptions of execution environments, required capabilities, and data movement restrictions directly

into their representation, fostering more efﬁcient and ﬂexible multi-facility collaborations [35].

Data movement across facilities remains a signiﬁcant challenge. While some institutions use Globus [

]

for data transfer and workﬂow automation, others rely on alternatives like Rsync or custom solutions due to

access limitations or security concerns. There is then a need for more universal and secure data movement

solutions, especially for sensitive data that cannot leave certain jurisdictions (e.g., health and defense data).

To address these multifaceted challenges, participants proposed a strategic, multi-tiered approach. The

immediate focus is on developing and showcasing demonstrable use cases, such as JAWS [

] and Stream-

Flow [

], supported by sustained funding for common solution development. Implementing multi-facility

federated AI enables private, high-performance LLM training, inspired by projects like EuroHPC [

Concurrently, the development of multi-facility-aware workﬂow provenance standards was identiﬁed as a

WORKFLOWS COMMUNITY SUMMIT 2024

major step towards enhancing reproducibility, performance portability, and debugging capabilities across

diverse environments [

]. Cross-Facility Federated Learning (XFFL) [

] integrates workﬂow abstractions

with Federated Learning, enabling cross-facility deep learning without large data transfers between sites.

Multi-facility workﬂows face signiﬁcant challenges, particularly in coordinating availability across multiple

HPCs and managing job failures due to maintenance. These issues can invalidate federated training efforts.

Developing robust fault tolerance mechanisms within new frameworks is crucial for ensuring successful

execution of cross-facility collaborations and advancing scientiﬁc research [41].

Future goals include developing domain-agnostic workﬂows and distributed runtime systems, moving

beyond centralized control/data plane orchestrators for greater ﬂexibility and scalability. A key aspect of this

strategy is the compilation of workﬂow models into distributed execution plans, leveraging multi-facility-

aware, low-level intermediate representations. Looking at this direction, the SWIRL language [

] shifts

the focus from high-level workﬂow languages, designed either for direct human interaction or for encoding

complex, product-speciﬁc features, towards a low-level minimalistic representation of a workﬂow execution

plan, making it more manageable for formalization methods and compiler toolchains. Last, discussions

explored integrating AI and quantum computing into multi-facility workﬂows, addressing AI’s unique

requirements (e.g., AI-at-the-edge workﬂows) and quantum computing’s need for classical fallback options

due to current device unreliability [43].

Recommendations:

•

Develop standardized cross-facility protocols for secure data movement, surpassing current

solutions while ensuring interoperability and maintaining strict security standards.

•

Implement distributed-by-design workﬂow modeling to create workﬂows inherently suited for

multi-facility execution, incorporating environment details and constraints.

•

Develop multi-facility-aware provenance standards and runtimes to compile workﬂows into

facility-speciﬁc plans, improving reproducibility and performance across diverse environments.

•

Explore and develop AI-ready data preparation solutions across facilities, addressing “dirty”

experimental data challenges and multi-facility AI workﬂow requirements.

•

Investigate quantum computing integration in multi-facility workﬂows, developing classical

simulation fallbacks for continuity and testing.

6 Heterogeneous HPC Environments

HPC is experiencing a radical shift towards increasingly heterogeneous environments. This transformation

is clearly reﬂected in the latest Top500 list [

], where eight of the top ten systems heavily rely on GPUs for

peak performance, and approximately 35% of all listed machines incorporate hardware accelerators. The trend

extends beyond traditional computing centers to encompass edge computing, creating a diverse computational

continuum from edge devices to HPC systems and potentially cloud environments. As scientiﬁc workﬂows

grow more sophisticated, often combining traditional HPC simulations with AI models and spanning multiple

facilities, ensuring performance portability across these diverse systems has become a critical challenge.

Emerging Paradigms and Challenges. Heterogeneous environments in HPC have become increasingly

prevalent and complex in the past years. The introduction of exascale systems like Frontier at Oak Ridge

National Laboratory and Aurora at Argonne National Laboratory, coupled with the pervasive inﬂuence of AI

and LLMs, the emergence of specialized AI hardware, and the integration of dedicated AI partitions in large

WORKFLOWS COMMUNITY SUMMIT 2024

supercomputers, have fundamentally reshaped the HPC ecosystem. These advancements have intensiﬁed

challenges in achieving performance portability across diverse hardware architectures, including CPUs,

GPUs from various vendors, and specialized AI accelerators. Users now face a daunting array of tools and

programming models, making it difﬁcult to navigate the heterogeneous landscape efﬁciently. The integration

of AI workﬂows with traditional HPC applications has created a convergence of two distinct computing

paradigms, each with unique resource requirements and optimization strategies. Moreover, energy costs and

system reliability have emerged as critical concerns in managing these intricate environments, adding further

complexity to workﬂow design and execution.

Recent Advances and Emerging Solutions. The HPC community is developing solutions for increasingly

heterogeneous computing environments. A key focus is integrating workﬂow considerations into early HPC

system design and deployment (e.g., including workﬂow benchmarks in the procurement process), creating

more workﬂow-friendly architectures. Efforts to standardize resource management and workﬂow tools are

ongoing, with projects like Flux [

] and libEnsemble [

] gaining prominence. These initiatives aim to

simplify user experience across systems and reduce the learning curve for heterogeneous resources. Research

into high-productivity, high-performance programming models is progressing, with languages like Julia

and Python-based solutions offering accessible interfaces while leveraging low-level optimizations. There

is also growing interest in harnessing AI and LLMs to assist with HPC programming and optimization

tasks. The community is exploring regular virtual meetings and standardized workﬂow test suites to foster

collaboration. These efforts aim to enhance portability, improve user training, and ensure effective utilization

of heterogeneous HPC environments across scientiﬁc domains.

Discussion. Heterogeneity in HPC environments encompasses a broad spectrum of contexts, including

hardware diversity (CPUs, GPUs, and specialized accelerators), federated systems (multi-cluster, multi-site,

facility-cloud integrations), varied applications and workﬂow tools, and data/storage heterogeneity. This

multifaceted nature of heterogeneity underscores the complexity of challenges facing the HPC community

and necessitates comprehensive solutions that address all aspects of heterogeneity in computing environments.

The proliferation of HPC workﬂow tools like Parsl [

], RADICAL Cybertools [

], libEnsemble,

PSI/J [

], Merlin [

], SmartSim [

], Dragon [

], AFCL [

], and StreamFlow [

], alongside numerous

homegrown solutions, underscores the need for better interoperability and common interfaces. These are

key for developing and executing portable workﬂows across heterogeneous resources. There is then growing

interest in developing a registry of execution environment information and standardizing resource description

formats to improve resource discovery and allocation.

Scheduling and resource allocation in heterogeneous environments remain critical challenges. The com-

plexities of both online (runtime) and ofﬂine (compile-time) scheduling underscore the need for sophisticated

algorithms capable of handling the diverse constraints and optimization metrics inherent in heterogeneous

systems [

]. AI and LLMs show promise in addressing these challenges, particularly in areas like automated

resource discovery, performance prediction, workﬂow optimization, and scheduling [55].

To accelerate progress in managing and leveraging heterogeneous HPC environments, several initiatives

are being considered. These include regular knowledge-sharing sessions, the development of comprehensive

documentation on heterogeneous packages, and the creation of standardized workﬂow test suites [

;

Such test suites could serve as benchmarks for evaluating different solutions and potentially aid in procurement

decisions and system testing. These efforts aim to foster collaboration and innovation, ultimately enabling

more efﬁcient and effective scientiﬁc workﬂows across diverse computing resources.

WORKFLOWS COMMUNITY SUMMIT 2024

Recommendations:

•

Create a central registry of execution environment information with standardized resource

descriptions to enable discovery, allocation, and portability across heterogeneous HPC systems.

• Develop adaptive scheduling algorithms for efﬁcient online and ofﬂine management in hetero-

geneous environments across diverse hardware, considering varied constraints and metrics.

•

Create community-driven workﬂow benchmarks and test cases spanning diverse scientiﬁc

domains and computational patterns to guide system evaluation, procurement, and optimization.

•

Integrate AI and machine learning techniques, especially LLMs, into HPC workﬂow systems to

automate resource discovery and optimization across heterogeneous computing environments.

7 User Experience and Interfaces

As scientiﬁc workﬂows continue to evolve in complexity and scale, the importance of User Experience

(UX) [

] and intuitive interfaces has become increasingly critical. Well-designed UX can signiﬁcantly

impact the adoption, efﬁciency, and overall success of workﬂow systems, empowering researchers to focus

on their scientiﬁc inquiries rather than grappling with technical intricacies. The interfaces (whether command

line, graphical, API, or documentation) between scientists and workﬂow systems serve as a key bridge,

determining how effectively researchers can harness the power of these tools.

Emerging Paradigms and Challenges. UX in scientiﬁc workﬂows has gained increased recognition in recent

years, with a broader understanding that UX encompasses more than just graphical user interfaces. The

concept now includes the entire user journey, from product discovery to adoption, and extends to the policies

and procedures governing services and facilities. This broadened perspective comes at a time of increasing

complexity in the workﬂow landscape, characterized by a proliferation of systems and interfaces that reﬂect

the growing heterogeneity of facilities and systems across national and global research infrastructures. This

expanding ecosystem presents signiﬁcant challenges for usability and interoperability among workﬂow

systems. The integration of AI and machine learning into scientiﬁc workﬂows has also introduced new data

curation challenges, affecting the adoption and integration of these tools. Current challenges include the need

for seamless interactivity in heterogeneous environments, reducing barriers to entry for workﬂow adoption,

and addressing the “closed box” issue where workﬂow processes remain opaque to end-users, hindering

troubleshooting and adaptability.

Recent Advances and Emerging Solutions. Workﬂow system developers and UX researchers are actively

working to simplify workﬂow usage, aiming to make it as intuitive as operating individual tools. Their efforts

focus on easing the transition for scientists from manual tool chaining to the adoption of comprehensive

workﬂow systems. Signiﬁcant progress has been made in developing improved graphical representations of

workﬂow steps and data ﬂow, enhancing transparency and user comprehension. The scientiﬁc community

is also adapting concepts from commercial software development, such as Design Systems, to the unique

needs of scientiﬁc applications. Initiatives like the Chan Zuckerberg Initiative’s open-source science design

system [

] and the STRUDEL project [

] are creating templates, code, and guidelines that enable teams

to craft consistent and user-friendly interfaces across different scientiﬁc domains. On the other hand, the

Jupyter Workﬂow [

] framework builds on the well-known Jupyter Notebook interface to let users design

and run interactive large-scale distributed workﬂows with sequential equivalence guarantees. These efforts

aim to improve the overall user experience of scientiﬁc workﬂows, from setup and conﬁguration to execution

WORKFLOWS COMMUNITY SUMMIT 2024

and troubleshooting, ultimately fostering better collaboration between humans, machines, and research

infrastructure systems.

Discussion. UX in scientiﬁc workﬂows encompasses a wide spectrum of perspectives and implementations.

Notable advancements in UX design have emerged, with some systems successfully integrating human

participation within their graphical user interfaces for tasks such as quality control and experiment tuning.

Industry-standard workﬂow platforms in bioinformatics, AI/ML, and cloud computing offer sophisticated UX

features that serve as benchmarks for the ﬁeld. However, replicating such interfaces in HPC environments

presents unique challenges due to the inherent complexity of these ecosystems and their specialized technical

and social requirements.

A central tension in workﬂow UX design lies between simplicity and complexity. There is a clear

need to make workﬂows more accessible to newcomers, yet many researchers value the ability to engage

with sophisticated tools as part of their learning and research processes. This dichotomy underscores the

importance of developing ﬂexible interfaces that can accommodate users with varying levels of expertise and

diverse workﬂow requirements, allowing for both ease of use and depth of functionality [62].

The potential role of AI, particularly LLMs, in enhancing workﬂow UX is a subject of ongoing debate

within the scientiﬁc community. While there is potential for AI to assist in generating workﬂow templates and

improving documentation, concerns persist regarding the interpretability and maintainability of AI-generated

workﬂows. These discussions highlight the need for thoughtful integration of AI technologies in scientiﬁc

workﬂow systems, ensuring that they enhance rather than complicate the user experience.

Heterogeneity and portability remain critical challenges in workﬂow UX. The scientiﬁc community

desires the ﬂexibility to run workﬂows across different systems while still accessing specialized features of

speciﬁc platforms. Successful approaches in addressing this challenge include developing systems that can

automatically identify their execution environment and apply appropriate conﬁgurations. However, this raises

fundamental questions about the optimal level of abstraction for workﬂows and how to balance the competing

demands of portability and platform-speciﬁc optimization.

Recommendations:

•

Formulate UX principles for scientiﬁc workﬂows, addressing common issues and guiding the

design of intuitive, user-friendly interfaces (CLIs, APIs, GUIs) across various workﬂow systems.

•

Create tailored onboarding and training for new workﬂow system users, offering hands-on

experience with real scientiﬁc problems and teaching core workﬂow management concepts.

•

Develop adaptive interfaces for workﬂow systems that accommodate users of varying expertise,

balancing simplicity for beginners with advanced features for experts.

•

Explore responsible AI integration (LLMs) in workﬂow systems for assisted generation, docu-

mentation, and troubleshooting, while preserving human interpretability and maintainability.

8 FAIR Computational Workﬂows

As scientiﬁc workﬂows become increasingly integral to data-intensive research across disciplines, ensur-

ing their ﬁndability, accessibility, interoperability, and reusability (FAIR) has emerged as a critical challenge

in the scientiﬁc community. The FAIR principles [

], originally developed for research data, and later

software [

] are now being adapted to computational workﬂows to enhance their discoverability, repro-

ducibility, and reuse. This shift is decisive as workﬂows embody not just data, but also scientiﬁc methods,

WORKFLOWS COMMUNITY SUMMIT 2024

analysis processes, and computational environments. By making workﬂows FAIR, researchers aim to improve

collaboration, accelerate scientiﬁc discovery, and increase the overall efﬁciency of computational research.

Emerging Paradigms and Challenges. The application of FAIR principles to computational workﬂows has

emerged as a critical focus in the scientiﬁc community [

]. As workﬂows become increasingly central

to data-intensive research across disciplines, ensuring their FAIRness presents unique challenges. Unlike

static datasets, workﬂows encompass both declarative descriptions and executable components, necessitating

a nuanced approach to FAIR implementation. The complexity is further compounded by the diversity of

workﬂow languages, execution environments, and domain-speciﬁc requirements. Emerging paradigms

in workﬂow design, such as modular and composable workﬂows, raise questions about how to maintain

FAIRness at various levels of granularity. Additionally, the integration of workﬂows with HPC environments

and cloud infrastructures introduces new considerations for accessibility and portability. Balancing the need

for standardization with the ﬂexibility required for innovation remains a key challenge.

Recent Advances and Emerging Solutions. Recent advances in making workﬂows FAIR have seen signiﬁcant

progress, with the development of FAIR principles speciﬁcally tailored to computational workﬂows [

These principles treat workﬂows as both data and software, addressing their unique dual nature. To support

practical implementation, the community has developed infrastructure and tools such as DockStore [

]

and WorkﬂowHub [

;

], which facilitate the registration and sharing of workﬂow metadata, thereby

enhancing ﬁndability and accessibility. Guidelines like “Ten quick tips for building FAIR workﬂows” [

]

provide researchers with actionable steps towards FAIRness. Efforts to improve interoperability between

workﬂow management systems are ongoing, with a focus on rich metadata descriptions and standardized

APIs. Emerging solutions also explore the potential of AI technologies in workﬂow programming and

automated metadata generation. However, challenges remain in effectively embedding FAIR practices into

workﬂow creation interfaces and extending the FAIR framework to encompass additional crucial aspects such

as portability, reproducibility, and quality assurance [71].

Within the Workﬂows Community Initiative, a working group has now established a set of principles

for FAIR Computational Workﬂows [

]. These built on the FAIR principles for data and for research

software [

], observed best practices, and considered workﬂows as an explainable composition of workﬂow

components, which themselves may be data sources, research software, or other workﬂows.

Discussion. The application of the FAIR principles to computational workﬂows has sparked signiﬁcant

discussions within the scientiﬁc community, revealing several key challenges and potential directions for

future development. A central topic is the deﬁnition and implementation of FAIR workﬂow components,

which necessitates engagement with the broader ecosystem surrounding workﬂows, including dependencies,

codes, and data. The challenge lies in striking a balance between comprehensive FAIRness and practicality,

while avoiding the risk of overcomplicating the process by attempting to make every aspect FAIR.

Implementing FAIR principles in practice remains a signiﬁcant hurdle in the workﬂow domain [

Poor annotation of workﬂows and difﬁculties in sharing workﬂows across different languages are common

issues. Proposed solutions include leveraging LLMs for workﬂow annotation, encouraging WMSs to adopt

shared models for workﬂow description (such as the Bioschemas Computational Workﬂow proﬁle [

]),

and promoting best practices among developers. The role of tools and services in encouraging FAIR

practices is crucial, with efforts to engage tool developers in implementing FAIR-supporting features, such

as the integration of Research Object Crate (RO-Crate) [

] speciﬁcations in WMSs. This integration is

demonstrated at [74], where six of these systems integrate the Workﬂow Run RO-Crate Proﬁle.

There is growing interest in developing a FAIR workﬂow maturity model, drawing parallels with software

WORKFLOWS COMMUNITY SUMMIT 2024

maturity models and certiﬁcation badges. Such a model could provide guidance for implementing FAIR

principles and offer a framework for assessing the FAIRness of workﬂows. The development of FAIR metrics

speciﬁcally tailored for workﬂows is seen as a key next step. Tools for FAIR assessment could automatically

provide a FAIRness score or grade, presenting which principles are being evaluated within the tests [

Tools like F-UJI [

] are being assessed for workﬂow applicability, with possible adaptations or new tools in

development. The community is also exploring the relationship between FAIR metrics and maturity models,

considering how they can complement each other in assessing and improving workﬂow FAIRness.

The application of FAIR principles to workﬂows containing export-controlled or conﬁdential components

presents unique challenges. This has highlighted the need for ﬂexible approaches to FAIRness that can

accommodate varying levels of openness and access restrictions. The concept of “Inner FAIR” within

organizations has emerged as a potential solution, allowing for the application of FAIR principles even in

contexts where full public sharing is not feasible. In areas in which sensitive data need to be managed, such as

human genomics, users need to go through data access request procedures that are often manually managed

by data access committees. This hinders the automation potential provided by computational workﬂow

technologies. Initiatives such as the Data Use Ontology (DUO) [

], that allows for formally expressing

the authorized uses for a dataset, could help to streamline tasks related to data discovery and retrieval

in computational workﬂows that need to process controlled-access data. Additionally, the community is

grappling with the distinction between reusability and reproducibility in the context of workﬂows, recognizing

that a workﬂow can be FAIR without necessarily being portable or reproducible across all environments.

Recommendations:

•

Promote standardized metadata schemas for workﬂows (Bioschemas ComputationalWorkﬂow

proﬁle, RO-Crate speciﬁcations) across WMSs to enhance interoperability and FAIR practices.

•

Develop a FAIR workﬂow maturity model and metrics to guide FAIR implementation, offering

standardized and automated assessment of workﬂow FAIRness, considering unique challenges

of components and execution environments.

•

Collaborate with tool developers to integrate FAIR-supporting features into workﬂow develop-

ment tools, including automatic rich metadata generation, annotation support (leveraging AI

technologies), and facilitation of workﬂow sharing across different languages and platforms.

•

Develop guidelines and tools for implementing “Inner FAIR” practices within organizations to

address the challenges of applying FAIR principles to workﬂows with conﬁdential or export-

controlled components, enabling FAIR practices even when full public sharing is not possible.

WORKFLOWS COMMUNITY SUMMIT 2024

References

[1]

J. Ejarque, R. M. Badia, L. Albertin, G. Aloisio, E. Baglione, Y. Becerra, S. Boschert, J. R. Berlin, A. D’Anca, D. Elia et al.,

“Enabling dynamic and intelligent workﬂows for HPC, data analytics, and AI convergence,” Future generation computer

systems, vol. 134, 2022. [Online]. Available: https://doi.org/10.1016/j.future.2022.04.014

[2]

W. Brewer, A. Gainaru, F. Suter, F. Wang, M. Emani, and S. Jha, “AI-coupled HPC Workﬂow Applications, Middleware and

Performance,” arXiv preprint arXiv:2406.14315, 2024. [Online]. Available: https://doi.org/10.48550/arXiv.2406.14315

[3]

R. Ferreira da Silva, R. M. Badia, D. Bard, I. T. Foster, S. Jha, and F. Suter, “Frontiers in Scientiﬁc Workﬂows: Pervasive

Integration with HPC,” IEEE Computer, vol. 57, no. 8, 2024. [Online]. Available: https://doi.org/10.1109/MC.2024.3401542

[4]

R. M. Badia, L. Berti-Equille, R. Ferreira da Silva, and U. Leser, “Integrating HPC, AI, and Workﬂows for Scientiﬁc Data

Analysis,” Dagstuhl Reports, vol. 13, no. 8, 2024. [Online]. Available: https://doi.org/10.4230/DagRep.13.8.129

[5]

K. B. Antypas, D. Bard, J. P. Blaschke, R. S. Canon, B. Enders, M. A. Shankar, S. Somnath, D. Stansberry, T. D. Uram,

and S. R. Wilkinson, “Enabling Discovery Data Science Through Cross-Facility Workﬂows,” in 2021 IEEE International

Conference on Big Data (Big Data). IEEE, 2021. [Online]. Available: https://doi.org/10.1109/BigData52589.2021.9671421

[6] “Workﬂows Community Summit 2024,” https://workﬂows.community/summits/2024, 2022.

[7] “Workﬂows Community Initiative,” https://workﬂows.community, 2024.

[8]

R. Francis and J. H. Christiansen, “Australian BioCommons Strategic Plan 2023 - 2028,” 2024. [Online]. Available:

https://doi.org/10.5281/zenodo.13626350

[9]

R. Ferreira da Silva, H. Casanova, K. Chard, I. Altintas, R. M. Badia, B. Balis, T. a. Coleman, F. Coppens,

F. Di Natale, B. Enders, T. Fahringer, R. Filgueira et al., “A Community Roadmap for Scientiﬁc Workﬂows Research and

Development,” in 2021 IEEE Workshop on Workﬂows in Support of Large-Scale Science (WORKS), 2021. [Online]. Available:

https://doi.org/10.1109/WORKS54523.2021.00016

[10]

R. Ferreira da Silva, R. M. Badia, V. Bala, D. Bard, T. Bremer, I. Buckley, S. Caino-Lores, K. Chard, C. Goble, S. Jha,

D. S. Katz, D. Laney, M. Parashar, F. Suter, N. Tyler, T. Uram, I. Altintas et al., “Workﬂows Community Summit 2022:

A Roadmap Revolution,” Oak Ridge National Laboratory, Tech. Rep. ORNL/TM-2023/2885, 2023. [Online]. Available:

https://doi.org/10.5281/zenodo.7750670

[11]

N. Brown, R. Nash, G. Gibb, E. Belikov, A. Podobas, W. Der Chien, S. Markidis, M. Flatken, and A. Gerndt, “Workﬂows to

driving high-performance interactive supercomputing for urgent decision making,” in International Conference on High

Performance Computing. Springer, 2022. [Online]. Available: https://doi.org/10.1007/978-3-031-23220-6 16

[12]

B. L. Brown, W. L. Miller et al., “Integrated Research Infrastructure Architecture Blueprint Activity,” Technical Report. US

Department of Energy (USDOE), Washington, DC, Tech. Rep., 2023. [Online]. Available: https://doi.org/10.2172/1984466

[13]

N. Wedi, P. Bauer, I. Sandu, J. Hoffmann, S. Sheridan, R. Cereceda, T. Quintino, D. Thiemert, and T. Geenen, “Destination

Earth: High-Performance Computing for Weather and Climate,” Computing in Science & Engineering, vol. 24, no. 6, 2022.

[Online]. Available: https://doi.org/10.1109/MCSE.2023.3260519

[14]

A. J. Koomthanam, S. Serebryakov, A. Tripathy, G. Nayak, M. Foltin, and S. Bhattacharya, “Common Metadata

Framework: Integrated Framework for Trustworthy AI Pipelines,” IEEE Internet Computing, 2024. [Online]. Available:

https://doi.org/10.1109/MIC.2024.3377170

[15]

S. Soiland-Reyes, P. Sefton, M. Crosas, L. J. Castro, F. Coppens, J. M. Fern

andez, D. Garijo, B. Gr

uning, M. La Rosa,

S. Leo et al., “Packaging Research Artefacts with RO-Crate,” Data Science, vol. 5, no. 2, 2022. [Online]. Available:

https://doi.org/10.3233/DS-210053

[16] “FAIR for Machine Learning (FAIR4ML) IG,” https://www.rd-alliance.org/groups/fair-machine-learning-fair4ml-ig, 2024.

[17]

E. Dart et al., “ESnet Requirements Review Program Through the IRI Lens: A Meta-Analysis of Workﬂow Patterns Across

DOE Ofﬁce of Science Programs,” Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States), Tech. Rep.,

2023. [Online]. Available: https://doi.org/10.2172/2008205

[18]

M. Graham, R. Milne, P. Fitzsimmons, and M. Sheehan, “Trust and the Goldacre Review: Why Trusted Research

Environments are not About Trust,” Journal of Medical Ethics, vol. 49, no. 10, 2023. [Online]. Available:

https://doi.org/10.1136/jme-2022-108435

[19]

C. Voisin, M. Linden, S. O. Dyke, S. R. Bowers, P. Alper, M. P. Barkley, D. Bernick, J. Chao, M. Courtot, F. Jeanson

et al., “GA4GH Passport Standard for Digital Identity and Access Permissions,” Cell Genomics, vol. 1, no. 2, 2021. [Online].

Available: https://doi.org/10.1016/j.xgen.2021.100030

[20]

T. Desai, F. Ritchie, R. Welpton et al., “Five Safes: Designing Data Access for Research,” Economics Working Paper Series,

vol. 1601, p. 28, 2016. [Online]. Available: https://doi.org/10.13140/RG.2.1.3661.1604

WORKFLOWS COMMUNITY SUMMIT 2024

[21]

L. Brescia and M. Aldinucci, “Secure generic remote workﬂow execution with TEEs,” in Proceedings of

the 2nd Workshop on Workﬂows in Distributed Environments, WiDE 2024. ACM, 2024. [Online]. Available:

https://doi.org/10.1145/3642978.3652834

[22]

L. Brescia, I. Colonnelli, and M. Aldinucci, “Performance analysis on DNA alignment workload with Intel SGX multithreading,”

in Proceedings of the 2nd Special Track on Big Data and High-Performance Computing (BigHPC 2024), ser. CEUR Workshop

Proceedings, vol. 3785. CEUR-WS.org, 2024. [Online]. Available: https://ceur-ws.org/Vol-3606/invited77.pdf

[23]

D. Manubens-Gil, J. Vegas-Regidor, C. Prodhomme, O. Mula-Valls, and F. J. Doblas-Reyes, “Seamless management of

ensemble climate prediction experiments on hpc platforms,” in 2016 International Conference on High Performance Computing

& Simulation (HPCS). IEEE, 2016.

[24]

L. Ward, J. G. Pauloski, V. Hayot-Sasson, Y. Babuji, A. Brace, R. Chard, K. Chard, R. Thakur, and I. Foster, “Employing

Artiﬁcial Intelligence to Steer Exascale Workﬂows with Colmena,” arXiv preprint arXiv:2408.14434, 2024. [Online]. Available:

https://doi.org/10.48550/arXiv.2408.14434

[25]

F. Suter, R. Ferreira da Silva, A. Gainaru, and S. Klasky, “Driving Next-Generation Workﬂows from the Data Plane,” in 19th

IEEE Conference on eScience, 2023. [Online]. Available: https://doi.org/10.1109/e-Science58273.2023.10254849

[26]

W. F. Godoy, N. Podhorszki, R. Wang, C. Atkins, G. Eisenhauer, J. Gu, P. E. Davis, J. Choi, K. Germaschewski, K. A. Huck,

A. Huebl, M. Kim et al., “ADIOS 2: The adaptable input output system. A framework for high-performance data management,”

SoftwareX, vol. 12, 2020. [Online]. Available: https://doi.org/10.1016/J.SOFTX.2020.100561

[27]

F. Li and F. Song, “INSTANT: A runtime framework to orchestrate in-situ workﬂows,” in Euro-Par 2023: Parallel Processing -

29th International Conference on Parallel and Distributed Computing, Limassol, Cyprus, August 28 - September 1, 2023,

Proceedings, vol. 14100. Springer, 2023. [Online]. Available: https://doi.org/10.1007/978-3-031-39698-4\ 14

[28]

M. Folk, A. Cheng, and K. Yates, “HDF5: A ﬁle format and I/O library for high performance computing applications,” in SC

’99: Proc. of the ACM/IEEE Conf. on Supercomputing, vol. 99, 1999.

[29]

O. Yildiz, D. Morozov, A. Nigmetov, B. Nicolae, and T. Peterka, “Wilkins: HPC in situ workﬂows made easy,” CoRR, vol.

abs/2404.03591, 2024. [Online]. Available: https://doi.org/10.48550/ARXIV.2404.03591

[30]

A. R. Martinelli, M. Torquati, M. Aldinucci, I. Colonnelli, and B. Cantalupo, “CAPIO: a middleware for

transparent I/O streaming in data-intensive workﬂows,” in 30th IEEE International Conference on High Performance

Computing, Data, and Analytics, HiPC 2023, Goa, India, December 18-21, 2023. IEEE, 2023. [Online]. Available:

https://doi.org/10.1109/HIPC58850.2023.00031

[31]

R. Souza, S. Caino-Lores, M. Coletti, T. J. Skluzacek, A. Costan, F. Suter, M. Mattoso, and R. F.

Da Silva, “Workﬂow Provenance in the Computing Continuum for Responsible, Trustworthy, and Energy-

Efﬁcient AI,” in 2024 IEEE 20th International Conference on e-Science (e-Science), 2024. [Online]. Available:

https://doi.org/10.1109/e-Science62913.2024.10678731

[32]

R. Souza, L. G. Azevedo, V. Louren

c¸

o, E. Soares, R. Thiago, R. Brand

ao, D. Civitarese, E. Vital Brazil, M. Moreno,

P. Valduriez, M. Mattoso, R. Cerqueira, and M. A. S. Netto, “Workﬂow provenance in the lifecycle of scientiﬁc

machine learning,” Concurrency and Computation: Practice and Experience, vol. e6544, 2021. [Online]. Available:

https://doi.org/10.1002/cpe.6544

[33]

B. Enders, D. Bard, C. Snavely, L. Gerhardt, J. Lee, B. Totzke, K. Antypas, S. Byna, R. Cheema, S. Cholia et al., “Cross-Facility

Science with the Superfacility Project at LBNL,” in 2020 IEEE/ACM 2nd Annual Workshop on Extreme-scale Experiment-in-

the-Loop Computing (XLOOP). IEEE, 2020. [Online]. Available: https://doi.org/10.1109/XLOOP51963.2020.00006

[34]

N. Tyler, R. Knop, D. Bard, and P. Nugent, “Cross-Facility Workﬂows: Case Studies with Active Experiments,” in 2022

IEEE/ACM Workshop on Workﬂows in Support of Large-Scale Science (WORKS). IEEE, 2022. [Online]. Available:

https://doi.org/10.1109/WORKS56498.2022.00014

[35]

I. Colonnelli, “Workﬂow models for heterogeneous distributed systems,” in Proceedings of the 2nd Italian

Conference on Big Data and Data Science (ITADATA 2023), vol. 3606. CEUR-WS.org, 2023. [Online]. Available:

https://ceur-ws.org/Vol-3606/invited77.pdf

[36]

K. Chard, I. Foster, and S. Tuecke, “Globus: Research Data Management as Service and Platform,” in Proceedings of the

Practice and Experience in Advanced Research Computing 2017 on Sustainability, Success and Impact, 2017. [Online].

Available: https://doi.org/10.1145/3093338.3093367

[37]

I. Colonnelli, B. Cantalupo, I. Merelli, and M. Aldinucci, “StreamFlow: cross-breeding cloud with HPC,” IEEE Transactions

on Emerging Topics in Computing, vol. 9, no. 4, 2021. [Online]. Available: https://doi.org/10.1109/TETC.2020.3019202

[38]

“The European High Performance Computing Joint Undertaking (EuroHPC JU),” https://eurohpc-ju.europa.eu/index en, 2024.

[39]

R. Souza, T. J. Skluzacek, S. R. Wilkinson, M. Ziatdinov, and R. Ferreira da Silva, “Towards lightweight data integration using

multi-workﬂow provenance and data observability,” in IEEE International Conference on e-Science, 2023. [Online]. Available:

https://doi.org/10.1109/e-Science58273.2023.10254822

WORKFLOWS COMMUNITY SUMMIT 2024

[40]

I. Colonnelli, R. Birke, G. Malenza, G. Mittone, A. Mulone, J. Galjaard, L. Y. Chen, S. Bassini, G. Scipione, J. Martinovi

V. Vondr

ak, and M. Aldinucci, “Cross-facility federated learning,” Procedia Computer Science, vol. 240, 2024. [Online].

Available: https://doi.org/10.1016/j.procs.2024.07.003

[41]

A. Mulone, D. Medi

c, and M. Aldinucci, “A fault tolerance mechanism for hybrid scientiﬁc workﬂows,” in Proceedings of 1st

workshop about High-Performance e-Science, HiPES, Madrid, Spain, 2024, 2024.

[42]

I. Colonnelli, D. Medi

c, A. Mulone, V. Bono, L. Padovani, and M. Aldinucci, “Introducing SWIRL: An intermediate

representation language for scientiﬁc workﬂows,” in Formal Methods. FM 2024, vol. 14933. Springer Nature Switzerland,

2024. [Online]. Available: https://doi.org/10.1007/978-3-031-71162-6\ 12

[43]

T. Beck, A. Baroni, R. Bennink, G. Buchs, E. A. Coello Perez, M. Eisenbach, R. Ferreira da Silva, M. Gopalakrishnan Meena,

K. Gottiparthi, P. Groszkowski, T. S. Humble, R. Landﬁeld, K. Maheshwari, S. Oral, M. A. Sandoval, A. Shehata, I.-S. Suh,

and C. Zimmer, “Integrating Quantum Computing Resources into Scientiﬁc HPC Ecosystems,” Future Generation Computer

Systems, vol. 161, pp. 11–25, 2024. [Online]. Available: https://doi.org/10.1016/j.future.2024.06.058

[44] “TOP500,” https://top500.org, 2024.

[45]

D. H. Ahn, N. Bass, A. Chu, J. Garlick, M. Grondona, S. Herbein, H. I. Ing

olfsson, J. Koning, T. Patki, T. R. Scogland et al.,

“Flux: Overcoming Scheduling Challenges for Exascale Workﬂows,” Future Generation Computer Systems, vol. 110, 2020.

[Online]. Available: https://doi.org/10.1109/WORKS.2018.00007

[46]

S. Hudson, J. Larson, J.-L. Navarro, and S. M. Wild, “libEnsemble: A Library to Coordinate the Concurrent Evaluation of

Dynamic Ensembles of Calculations,” IEEE Transactions on Parallel and Distributed Systems, vol. 33, no. 4, 2021. [Online].

Available: https://doi.org/10.1109/TPDS.2021.3082815

[47]

Y. Babuji, A. Woodard, Z. Li, D. S. Katz, B. Clifford, R. Kumar, L. Lacinski, R. Chard, J. M. Wozniak, I. Foster et al.,

“Parsl: Pervasive Parallel Programming in Python,” in Proceedings of the 28th International Symposium on High-Performance

Parallel and Distributed Computing, 2019. [Online]. Available: https://doi.org/10.1145/3307681.3325400

[48]

M. Turilli, V. Balasubramanian, A. Merzky, I. Paraskevakos, and S. Jha, “Middleware Building Blocks for Workﬂow Systems,”

Computing in Science & Engineering, vol. 21, no. 4, 2019. [Online]. Available: https://doi.org/10.1109/MCSE.2019.2920048

[49]

M. Hategan-Marandiuc, A. Merzky, N. Collier, K. Maheshwari, J. Ozik, M. Turilli, A. Wilke, J. M. Wozniak, K. Chard, I. Foster,

R. Ferreira da Silva, S. Jha, and D. Laney, “PSI/J: A Portable Interface for Submitting, Monitoring, and Managing Jobs,” in

19th IEEE Conference on eScience, 2023. [Online]. Available: https://doi.org/10.1109/e-Science58273.2023.10254912

[50]

J. L. Peterson, B. Bay, J. Koning, P. Robinson, J. Semler, J. White, R. Anirudh, K. Athey, P.-T. Bremer, F. Di Natale et al.,

“Enabling Machine Learning-Ready HPC Ensembles with Merlin,” Future Generation Computer Systems, vol. 131, 2022.

[Online]. Available: https://doi.org/10.1016/j.future.2022.01.024

[51]

S. Partee, M. Ellis, A. Rigazzi, A. E. Shao, S. Bachman, G. Marques, and B. Robbins, “Using Machine Learning at Scale in

Numerical Simulations with SmartSim: An Application to Ocean Climate Modeling,” Journal of Computational Science,

vol. 62, 2022. [Online]. Available: https://doi.org/10.1016/j.jocs.2022.101707

[52]

N. Radcliffe, K. Lee, and P. Mendygral, “Dragon Proxy Runtimes and Multi-system Workﬂows,” in Proceedings of the SC’23

Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis, 2023. [Online].

Available: https://doi.org/10.1145/3624062.3624139

[53]

S. Ristov, S. Pedratscher, and T. Fahringer, “AFCL: An Abstract Function Choreography Language for

Serverless Workﬂow Speciﬁcation,” Future Generation Computer Systems, vol. 114, 2021. [Online]. Available:

https://doi.org/10.1016/j.future.2020.08.012

[54]

J. R. Coleman, R. V. Agrawal, E. Hirani, and B. Krishnamachari, “Parameterized task graph scheduling algorithm for comparing

algorithmic components,” CoRR, vol. abs/2403.07112, 2024. [Online]. Available: https://doi.org/10.48550/arXiv.2403.07112

[55]

M. Kiamari and B. Krishnamachari, “Gcnscheduler: scheduling distributed computing applications using graph convolutional

networks,” in Proceedings of the 1st International Workshop on Graph Neural Networking, GNNet 2022, Rome, Italy, 9

December 2022. ACM, 2022. [Online]. Available: https://doi.org/10.1145/3565473.3569185

[56]

T. Coleman, H. Casanova, K. Maheshwari, L. Pottier, S. R. Wilkinson, J. Wozniak, F. Suter, M. Shankar, and R. Ferreira da

Silva, “WfBench: Automated Generation of Scientiﬁc Workﬂow Benchmarks,” in 2022 IEEE/ACM International Workshop on

Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS), 2022. [Online].

Available: https://doi.org/10.1109/PMBS56514.2022.00014

[57]

D. Bard, T. Groves, B. Cook, L. Stephey, W. Bhimji, S. Farrell, B. Austin, K. Gott, S. Canon, K. Kallback-Rose, J. Srini-

vasan, H. A. Nam, and N. J. Wright, “Workﬂow Archetypes White Paper,” https://www.nersc.gov/assets/NERSC-10/

Workﬂows-Archetypes-White-Paper-v1.0.pdf, National Energy Research Scientiﬁc Computing Center, Tech. Rep., 2023.

[58]

D. Paine, “Framing user experience (ux) across the scientiﬁc software lifecycle,” https://bssw.io/blog posts/

framing-user-experience-ux-across-the-scientiﬁc-software-lifecycle, 2024.

[59] “Essential Open Source Software for Science,” https://chanzuckerberg.com/eoss, 2024.

WORKFLOWS COMMUNITY SUMMIT 2024

[60]

“STRUDEL: Scientiﬁc sofTware Research for User experience, Design, Engagement, and Learning,” https://strudel.science/,

2024.

[61]

I. Colonnelli, M. Aldinucci, B. Cantalupo, L. Padovani, S. Rabellino, C. Spampinato, R. Morelli, R. D. Carlo, N. Magini,

and C. Cavazzoni, “Distributed workﬂows with Jupyter,” Future Gener. Comput. Syst., vol. 128, 2022. [Online]. Available:

https://doi.org/10.1016/J.FUTURE.2021.10.007

[62]

D. Paine, S. Poon, and L. Ramakrishnan, “Investigating User Experiences with Data Abstractions on High Performance

Computing Systems,” https://doi.org/10.2172/1805039, Lawrence Berkeley National Laboratory, Tech. Rep., 2021.

[63]

M. D. Wilkinson, M. Dumontier et al., “The FAIR Guiding Principles for Scientiﬁc Data Management and Stewardship,”

Scientiﬁc Data, vol. 3, no. 1, 2016. [Online]. Available: https://doi.org/10.1038/sdata.2016.18

[64]

M. Barker, N. P. Chue Hong, D. S. Katz, A.-L. Lamprecht, C. Martinez-Ortiz, F. Psomopoulos, J. Harrow, L. J. Castro,

M. Gruenpeter, P. A. Martinez, and T. Honeyman, “Introducing the FAIR principles for research software,” Scientiﬁc Data,

vol. 9, no. 1, 2022. [Online]. Available: https://doi.org/10.1038/s41597-022-01710-x

[65]

C. Goble, S. Cohen-Boulakia, S. Soiland-Reyes, D. Garijo, Y. Gil, M. R. Crusoe, K. Peters, and D. Schober,

“FAIR Computational Workﬂows,” Data Intelligence, vol. 2, no. 1–2, p. 108–121, 2020. [Online]. Available:

https://doi.org/10.1162/dint a 00033

[66]

S. R. Wilkinson, M. Aloqalaa, K. Belhajjame, M. R. Crusoe, B. de Paula Kinoshita, L. Gadelha, D. Garijo, O. J. R. Gustafsson,

N. Juty, S. Kanwal, F. Z. Khan, J. K

oster, K. P. von Gehlen, L. Pouchard, R. K. Rannow, S. Soiland-Reyes, N. Soranzo, S. Suﬁ,

Z. Sun, B. Vilne, M. A. Wouters, D. Yuen, and C. Goble, “Applying the FAIR principles to Computational Workﬂows,” arXiv,

2024. [Online]. Available: https://doi.org/10.48550/arXiv.2410.03490

[67]

D. Yuen, L. Cabansay, A. Duncan, G. Luu, G. Hogue, C. Overbeck, N. Perez, W. Shands, D. Steinberg, C. Reid et al., “The

Dockstore: Enhancing a Community Platform for Sharing Reproducible and Accessible Computational Protocols,” Nucleic

Acids Research, vol. 49, no. W1, 2021. [Online]. Available: https://doi.org/10.1093/nar/gkab346

[68]

C. Goble, S. Soiland-Reyes, F. Bacall, S. Owen, A. Williams, I. Eguinoa, B. Droesbeke, S. Leo, L. Pireddu, L. Rodr

ıguez-Navas

et al., “Implementing FAIR Digital Objects in the EOSC-Life Workﬂow Collaboratory,” Zenodo, 2021. [Online]. Available:

https://doi.org/10.5281/zenodo.4605654

[69]

O. J. R. Gustafsson, S. R. Wilkinson, F. Bacall, S. Soiland-Reyes, S. Leo, L. Pireddu, S. Owen, N. Juty, J. M. Fern

andez,

T. Brown, H. M

enager, B. Gr

uning, S. Capella-Gutierrez, F. Coppens, and C. Goble, “WorkﬂowHub: a registry for

computational workﬂows,” arXiv, 2024. [Online]. Available: https://doi.org/10.48550/arXiv.2410.06941

[70]

C. de Visser, L. F. Johansson, P. Kulkarni, H. Mei, P. Neerincx, K. Joeri van der Velde, P. Horvatovich, A. J. van Gool, M. A.

Swertz, P. Hoen et al., “Ten Quick Tips for Building FAIR Workﬂows,” PLoS Computational Biology, vol. 19, no. 9, 2023.

[Online]. Available: https://doi.org/10.1371/journal.pcbi.1011369

[71]

S. Kanwal, F. Z. Khan, A. Lonie, and R. O. Sinnott, “Investigating reproducibility and tracking provenance–a genomic workﬂow

case study,” BMC bioinformatics, vol. 18, 2017.

[72]

S. R. Wilkinson, G. Eisenhauer, A. J. Kapadia, K. Knight, J. Logan, P. Widener, and M. Wolf, “F*** workﬂows: when parts of

FAIR are missing,” in 2022 IEEE 18th International Conference on e-Science (e-Science). IEEE, 2022. [Online]. Available:

https://doi.org/10.1109/eScience55777.2022.00090

[73]

“Bioschemas ComputationalWorkﬂow Proﬁle,” https://bioschemas.org/proﬁles/ComputationalWorkﬂow/1.0-RELEASE, 2021.

[74]

S. Leo, M. R. Crusoe, L. Rodr

ıguez-Navas, R. Sirvent, A. Kanitz, P. De Geest, R. Wittner, L. Pireddu, D. Garijo, J. M.

Fern

andez et al., “Recording provenance of workﬂow runs with RO-Crate,” PLoS one, vol. 19, no. 9, 2024. [Online]. Available:

https://doi.org/10.1371/journal.pone.0309210

[75]

L. G. Azevedo, G. Banaggia, J. Tesolin, and R. Cerqueira, “An appraisal of automated tools for fairness evaluation,” in 4th

Workshop on Metadata and Research Objects Management for Linked Open Science (DaMaLOS 2024), 2024. [Online].

Available: https://repository.publisso.de/resource/frl:6483276

[76]

R. Huber and A. Devaraju, “F-UJI: an Automated Tool for the Assessment and Improvement of the FAIRness of Research Data,”

in EGU General Assembly Conference Abstracts, 2021. [Online]. Available: https://doi.org/10.5194/egusphere-egu21-15922

[77]

J. Lawson, M. N. Cabili, G. Kerry, T. Boughtwood, A. Thorogood, P. Alper, S. R. Bowers, R. R. Boyles et al., “The Data Use

Ontology to Streamline Responsible Access to Human Biomedical Datasets,” Cell Genomics, vol. 1, no. 2, 2021. [Online].

Available: https://doi.org/10.1016/j.xgen.2021.100028

WORKFLOWS COMMUNITY SUMMIT 2024

Appendix A: Participants and Contributors

First Name Last Name Afﬁliation

Abel Souza UMass Amherst

Alberto Mulone University of Turin

Amal Gueroudji Argonne National Laboratory

Anderson Andrei Da Silva Hewlett Packard Enterprise Labs

Andy Gallo GE Aerospace Research

Barry Sly-Delgado University of Notre Dame

Ben Tovar University of Notre Dame

Billy Rowell Paciﬁc Biosciences

Brian Etz Oak Ridge National Laboratory

Bruno P. Kinoshita Barcelona Supercomputing Center

Carlo Hamalainen Standard Chartered

Carole Goble The University of Manchester

Chen Li UC Irvine

Daniel Balouek Inria

Daniel de Oliveira Universidade Federal Fluminense

Daniel Garijo Universidad Polit

ecnica de Madrid

Daniel Rosendo Oak Ridge National Laboratory

Daniela Cassol Lawrence Berkeley National Laboratory

Deborah Bard National Energy Research Scientiﬁc Computing Center

Dinindu Senanayake New Zealand eScience Infrastructure

Domenico Talia Universit

a della Calabria

Doriana Medi

c University of Turin

Drew Paine Lawrence Berkeley National Laboratory

Edite Vartina Riga Stradin¸

s University

Eric Vardar-Irrgang Lawrence Livermore National Laboratory

Fabian Lehmann Humboldt-Universit

at zu Berlin

Fred Suter Oak Ridge National Laboratory

Hector Sanchez Universidad de Puerto Rico

Henri Casanova University of Hawaii at Manoa

Iacopo Colonnelli University of Turin

Ian Foster Argonne National Laboratory

Jack Marquez University of Tennessee

Jacob Fosso Tande North Carolina State University

Jakob Luettgau Inria

Jan Janssen Max Planck Institute for Sustainable Materials

Jared Coleman Loyola Marymount University

Jedrzej Rybicki Forschungszentrum Juelich

Joe Jordan Uppsala University

Johan Gustafsson Australia BioCommons

John-Luke Navarro Argonne National Laboratory

Jonathan Bader TU Berlin

Jose M. Fernandez Barcelona Supercomputing Center

Karan Vahi USC Information Sciences Institute

Ketan Maheshwari Oak Ridge National Laboratory

Kevin Hunter Kesling University of Chicago

Khalid Belhajjame Universit

e Paris-Dauphine

Kin Wai Ng University of Tennessee

Klaus Noelp University of Hagen

Kyle Chard University of Chicago

Laila Los University of Freiburg

Lauren Huet University of Western Australia

Lauritz Thamsen University of Glasgow

Liliane Kunstmann UFRJ/COPPE

Logan Ward Argonne National Laboratory

Luiz Gadelha German Cancer Research Center

WORKFLOWS COMMUNITY SUMMIT 2024

Mahnoor Zulﬁqar EMBL Heidelberg

Marcos Amaris UFPA

Merridee Wouters UNSW Sydney

Michael R. Crusoe Common Workﬂow Language project

Mikhail Titov Brookhaven National Laboratory

Motohiko Matsuda RIKEN R-CCS

Nathan Tallent Paciﬁc Northwest National Laboratory

Nishant Saurabh Utrecht University

Nour Elfaramawy Humboldt University of Berlin

Pascal Elahi Pawsey Supercomputing Research Centre

Patricia Grubel Los Alamos National Laboratory

Paul Brunk University of Georgia

Paula Iborra Barcelona Supercomputing Center

Peini Liu Barcelona Supercomputing Center

Philipp Gritsch University of Innsbruck

Qi Yu NIH

Quentin Guilloteau University of Basel

Rafael Ferreira da Silva Oak Ridge National Laboratory

Rajat Bhattarai Tennessee Tech University

ul Sirvent Barcelona Supercomputing Center

Renan Souza Oak Ridge National Laboratory

Riccardo Balin Argonne National Lab

Richard Lupat Peter MacCallum Cancer Centre

Rolando Hong Enriquez Hewlett Packard Labs

Rosa Filgueira EPCC, University of Edinburgh

Ryan Prout Oak Ridge National Laboratory

Sarah Beecroft Pawsey Supercomputing Research Centre

Sashko Ristov University of Innsbruck

Sean Wilkinson Oak Ridge National Laboratory

Sehrish Kanwal University of Melbourne

Shantenu Jha Princeton Plasma Physics Laboratory

Shaun de Witt UK Atomic Energy Authority

Shiva Jahangiri Santa Clara University

Silvina Caino-Lores Inria

Somayeh Mohammadi M

alardalen University

Stefan Robila Montclair State University

Stephen Hudson Argonne National Lab

Stian Soiland-Reyes The University of Manchester

Sumit Kumar Saurav C-DAC

Taina Coleman San Diego Supercomputing Center

Tanu Malik DePaul University

Thomas Fahringer University of Innsbruck

Tom Gibbs Nvidia

Tom Scogland Lawrence Livermore National Laboratory

Tyler Skluzacek Oak Ridge National Laboratory

Ulf leser Humboldt-Universit

at zu Berlin

Utz-Uwe Haus Hewlett Packard Enterprise Labs

Wael Elwasif Oak Ridge National Laboratory

Wes Brewer Oak Ridge National Laboratory

Wesley Ferreira Universidade Federal Fluminense

William F. Godoy Oak Ridge National Laboratory

Woong Shin Oak Ridge National Laboratory

Yadu Babuji University of Chicago

Yiannis Georgiou Ryax Technologies

Yuandou Wang University of Amsterdam

Ziad Al Bkhetan University of Melbourne

WORKFLOWS COMMUNITY SUMMIT 2024

Appendix B: Agenda

Day 1 – September 3, 2024

11:00-11:10am EDT

Welcome and Introductions

11:10-11:20am EDT

Frontiers in Scientiﬁc Workﬂows: Pervasive Integration with HPC

Rafael Ferreira da Silva (Oak Ridge National Laboratory)

11:20-11:50am EDT Lightning talks

• Topic 1: Time-Sensitive Workﬂows

Utz-Uwe Haus (HPE), Shaun DeWitt (UKAEA)

• Topic 2: Convergence of AI and HPC Workﬂows

Shantenu Jha (PPPL), Logan Ward (ANL)

• Topic 3: Multi-Facility Workﬂows in Next-Generation Scientiﬁc Collaboration

Tom Gibbs (NVIDIA), Debbie Bard (LBNL)

11:50am-noon EDT 10min Break

noon-1:30pm EST Breakout sessions

1:30-1:45pm EDT 15min Break

1:45-2:00pm EDT Reports from breakout sessions

Day 2 – September 4, 2024

11:00-11:10am EDT

Summary and Highlights of Day 1 – Stephen Hudson (ANL)

11:10-11:45am EDT Lightning talks

• Topic 4: Heterogeneous HPC Environments

William Godoy (ORNL), Stephen Hudson (ANL)

• Topic 5: User Experience and Interfaces

Drew Paine (LBNL), Laila Los (UFreiburg)

• Topic 6: FAIR Computational Workﬂows

Carole Goble (UManchester), Sean Wilkinson (ORNL)

11:45am-noon EDT 15min Break

noon-12:20pm EDT

Invited talk: AI-mediated Scientiﬁc Workﬂows – Ian T. Foster (ANL/UChicago)

12:15-1:45pm EDT Breakout sessions

1:45-2:00pm EDT Reports from breakout sessions

Day 3 – September 5, 2024 (Special Event for Asia/Oceania timezones)

12:30-12:35 AEST

Welcome and Introductions

Sean Wilkinson (ORNL), Johan Gustafsson (Australian BioCommons)

12:35-12:45 AEST

Context setting presentation for the WCI

Sean Wilkinson (ORNL)

12:45-12:55 AEST

Perspective on Bioinformatics Workﬂows in Australia

Johan Gustafsson (Australian BioCommons, University of Melbourne)

12:55-13:35 AEST Lightning talks

• Clinical Genomics Workﬂows

Sehrish Kanwal (University of Melbourne)

• Heterogeneous HPC Environments

Stephen Hudson (ANL)

• Workﬂow Execution with Seqera Platform

Ziad Al Bkhetan (Australian BioCommons)

• Migrating Complex Workﬂows to the Exascale: Challenges for Radio Astronomy

Pascal Elahi (Pawsey, Kensington WA)

13:35-13:40 AEST 5min Break

13:40-14:40 AEST Breakout sessions

• Multi-Facility Workﬂows

• What makes a Workﬂow Good?

• Heterogeneous HPC Environments

14:40-15:00 AEST Reports from breakout sessions

Workflows Community Summit 2024

Future Trends and Challenges in Scientific Workflows

September 3–5, 2024

ORNL/TM-2024/3573

https://workflows.community/summits/2024/